Introducing Agent Memory on Databricks
Databricks ships managed agent memory on Agent Bricks, putting persistent state inside the agent platform.
AI Daily Recap
44 articles · 5 categories
The finishable daily brief
What happened in AI — Oct 20, 2026
Tuesday, Oct 20, 2026
44 articles · 5 categories
read top to bottom · then stop
In 30 seconds
- DeepSeek open-sources its large-scale inference stack to cut serving cost.
- U.S. bans DeepSeek from federal civilian IT systems; China objects.
- Databricks launches managed agent memory on Agent Bricks.
- llm-d 1.0 and vLLM TorchTPU advance open inference serving.
- Claude Opus 5.5 and Sonnet 5.5 arrive in Google Antigravity.
DeepSeek open-sourced the inference software behind its models on the same day the U.S. banned its products from federal systems.
Agent memory and self-verification tooling shipped from Databricks and several open-source projects.
Agent memory and verification tooling 5 items
Memory and self-verification moved from demos to shipped products and open-source tools.
Show HN: Skillsync – Coding agent memory but it's just search
Skillsync searches past sessions across harnesses on every prompt, treating agent memory as retrieval.
Should agent-memory benchmarks test stale information?
Asks whether agent-memory benchmarks should test stale information, a gap in current evals.
Show HN: Make AI coding agents verify its own work
Reticle open-sources a way to make coding agents check their own work before finishing.
Show HN: Gauge CLI – Open-Source Hillclimbing for Agent Experience
Gauge CLI open-sources hillclimbing of agent-facing surfaces such as skills and CLIs.
Evals for agents in production 2 items
Teams are publishing concrete eval methods for deciding whether an agent is ready to ship.
How To Use Decision Models for Evals
Hamel Husain shows how small decision models can score evals cheaply, following releases from several vendors.
Show HN: A practical framework for evaluating agentic AI systems
deepsense.ai lays out a framework for judging whether an agent that passed its eval is safe to ship.
Inference and serving infrastructure 5 items
Open inference stacks are cutting serving cost; DeepSeek's release is the day's largest.
DeepSeek Unveils Open-Source Tool to Help Cut AI Operating Costs
DeepSeek open-sources the inference software it runs its models on, aimed at lowering operating cost.
DeepSeek Dials Up the Pressure on OpenAI and Anthropic With Open-Source Inference Tool
Press coverage frames the release as pressure on OpenAI and Anthropic serving economics.
llm-d 1.0 Advances Open, Cost-Efficient AI Inference for Production Scale
llm-d 1.0 reaches release as an open stack for cost-efficient production inference.
Introducing vLLM TorchTPU
vLLM adds a TPU backend on TorchTPU with Pallas kernels, multi-host parallelism and KV cache offloading.
Presentation: Beyond the Prototype: Scaling Framework Agnostic AI Agent Infrastructure with Ray
InfoQ talk on moving agents from FastAPI prototypes to resilient Kubernetes production with Ray.
U.S. DeepSeek ban in federal systems 4 items
A federal ban on DeepSeek products changes vendor-risk calculations for teams selling into or working with U.S. agencies.
U.S. government bans Chinese AI app DeepSeek from federal IT systems
The U.S. bars DeepSeek's app from federal IT systems.
DeepSeek set to be banned from federal civilian agencies
FedScoop reports the ban covers federal civilian agencies.
U.S. Bans DeepSeek From Federal Agencies, and Windows Could Be Caught Up in It
Windows-related exposure is flagged as a possible knock-on effect of the ban.
China Condemns U.S. Ban on DeepSeek A.I. in Government Systems
China condemns the ban, signaling the dispute will continue.
Models and agent platforms 2 items
Frontier models and agent connectors keep spreading into the platforms builders already use.
Claude Opus 5.5 and Claude Sonnet 5.5 are now available in Google Antigravity
Claude Opus 5.5 and Sonnet 5.5 become selectable in Google Antigravity.
Work management context for AI agents with Smartsheet MCP on AWS
AWS shows two ways to give agents live Smartsheet data through its MCP server.
You are caught up for this edition